NEWS & VIEWS

nature materials | VOL 6 | JULY 2007 | www.nature.com/naturematerials 473

the symmetry further. Th e authors cleverly overcome this by incorporating islands of InAs inside their Mn-doped GaAs fi lm. Th e GaAs distorts around the larger lattice constant InAs, resulting in local strains that are both large and direction-dependent. In this way, Mn acceptors in widely diff ering strain environments could be investigated within a single sample, or even within a single STM image.

Yakunin et al. found that for Mn located directly above the InAs island, such that the strain is directed along the growth direction (the [001] crystalline axis, see Fig. 1), the symmetry of the hole is indistinguishable from that of the unstrained case (Fig. 1b). However, when the Mn atom is located to one side of the island, the off -axis strain fi eld gives rise to a dramatic distortion, and symmetry-lowering, of the hole state (Fig. 1c).

Calculations of the local electronic structure around the Mn atoms give good agreement with the observed shape. Th ese calculations provide a bridge between the atomic-scale observations and the macroscopic properties of the doped semiconductor, as the warping of the hole

state corresponds to a change in the magnetic anisotropy. Th e calculated dependence of the preferred Mn spin direction on strain is the same as that observed in ferromagnetic Ga1–xMnxAs fi lms with low hole concentrations. Th is suggests that the striking magnetic anisotropies that are observed in Ga1–xMnxAs at low hole doping6 may be related to the distortion of the bound hole.

Yakunin and colleagues’ experiments in ref. 1 were performed for low doping levels, that is, for a low concentration of Mn atoms, which can therefore be considered as isolated and non-interacting. A diff erent picture emerges for highly doped, conducting Ga1–xMnxAs fi lms, where the strain has the opposite eff ect on the preferred spin direction with respect to that derived by Yakunin et al. for a single acceptor. At high doping, hole orbitals from adjacent acceptors overlap signifi cantly, and the magnetic properties are well-described by a model of ferromagnetism mediated by fully itinerant valence-band holes2,3. However, measurements by Yakunin et al. on pairs of Mn acceptors7 indicate that the hole retains its anisotropic shape, even for very short

spatial separations of the pairs. Th is implies that the holes retain some of their ‘bound’ character even in highly doped Ga1–xMnxAs, and the extent to which this infl uences its magnetic properties is yet to be established.

Th e cross-sectional STM studies of Yakunin et al. present us with some intriguing new insights into the properties of acceptors in semiconductors. Combined with recent demonstrations8 of STM-assisted atom-by-atom incorporation of Mn into GaAs, the ability to apply large local strains suggests new methods for manipulating single spins in semiconductors. Th ey also confi rm that eff ort is still required before ferromagnetism in Mn-doped semiconductors can be considered to be fully understood.

References1. Yakunin, A. M. et al. Nature Mater. 6, 512–515 (2007).2. Dietl, T., Ohno, H. & Matsukura, F. Phys. Rev. B 63, 195205 (2001).3. Jungwirth, T. et al. Phys. Rev. B 72, 165204 (2005).4. Shen, A. et al. J. Cryst. Growth 175–176, 1069–1074 (1997).5. Yakunin, A. M. et al. Phys. Rev. Lett. 92, 216806 (2004).6. Sawicki, M. et al. Phys. Rev. B 70, 245325 (2004).7. Yakunin, A. M. et al. Phys. Rev. Lett. 95, 256402 (2005).8. Kitchen, D., Richardella, A., Tang, J. M., Flatte, M. E. &

Yazdani, A. Nature 442, 436–439 (2006).

OXIDE INTERFACES

Watch out for the lack of oxygen

James N. Ecksteinis in the Department of Physics, University of Illinois at Urbana-Champaign, 1110 West Green Street, Urbana, Ilinois 61801-3080, USA.

e-mail: eckstein@uiuc.edu

In early 2004, Ohtomo and Hwang reported that a very conducting two-dimensional electron-transport channel

develops at the interface between the two perovskite oxides LaAlO3 (LAO) and SrTiO3 (STO)1. Th e result generated considerable surprise because each of the two compounds is a strong electric insulator. Yet, when combined in a heterostrucuture they generate a conductive channel with transport properties superior to more conventional two-dimensional electron gases, for example those formed by GaAs/AlAs heterostructures. In particular, the carrier density was extremely high (of the order of 4 × 1016 cm–2). Th is unexpected result,

and the potential for the development of high-performance oxide-based electronic devices, has generated considerable eff ort to understand where this huge carrier density comes from. Th e details of the electronic interfaces become even more important because of the observation reported on page 493 of this issue by Brinkman et al. that the two layers, which besides being insulating are non-magnetic, can give rise to magnetic eff ects at the interface2. Th ree recent publications have reported experimental results that suggest that oxygen vacancies play a role in the high carrier concentrations observed, although their results diff er in important ways3–5.

To understand why the question arises at all, we need to analyse the structure of the heterojunction at the atomic level. Figure 1 shows a perovskite unit cell, which has the composition ABO3. Th e LAO structure is made of alternating planes of LaO and

AlO2, and STO is made up of alternating SrO and TiO2 planes. Both compounds are electrically neutral, but in LAO the LaO planes are charged +1, and the AlO2 planes are charged –1, whereas in STO both planes are uncharged. An n-type junction between these two materials is made by growing LaO on top of a TiO2-terminated surface. Th en the electric fi eld lines that connect planes of opposite charge in the LAO all point upwards as shown in Fig. 2. Th is is caused by the compositional boundary condition that is determined by the atomic structure of the interface. As the crystal grows, the chemical potential for electrons in the surface layer increases because of the unidirectional electric fi eld. Eventually, this has to stop, and a mechanism by which it can is to transfer negative charge to the interfacial TiO2 layer to avoid this ‘polarization catastrophe’. Th is may be accompanied by oxygen vacancies in the surface layer. If 0.5 of an electron per unit

Interfaces between certain insulating perovskite oxides show unexpected properties, such as high conductivity and magnetism. Oxygen vacancies seem to be important in these structures, but the puzzle is far from being understood.

nmat July N&V.indd 473nmat July N&V.indd 473 11/6/07 16:04:1611/6/07 16:04:16

NEWS & VIEWS

474 nature materials | VOL 6 | JULY 2007 | www.nature.com/naturematerials

cell accumulates in the interfacial TiO2 plane, the catastrophe is avoided. But this amounts to only 3.5 × 1014 cm–2, which is two orders of magnitude lower than the sheet density originally reported.

Th e atomic picture described above is based on perfect stoichiometry of the compounds. Recent reports, however, propose a small lack of oxygen as the origin of extra carriers that give rise to the high sheet density. Kalabukhov and collaborators systematically changed growth and annealing conditions and were able to tune the carrier density from over 4 × 1016 cm–2 down to 1013 cm–2 by adding more oxygen aft er growth. Th eir conclusion is that the high carrier density is induced by oxygen vacancies in the STO layer3. Herranz and co-workers suggest a more provocative explanation: they measured the conductance as a function of magnetic fi eld, and observe Shubnikov de Haas oscillations whose amplitude and frequency indicate a 3D electron distribution that extends hundreds of micrometres into the substrate, putting the role of the interface as the generator or even location of carriers in question4. On the other hand, Siemons and collaborators infer a much higher local charge density, ~7 × 1020 cm–3, at the interface to the LAO layer by measuring the ultraviolet photoemission spectrum obtained through a single unit cell of LAO on top of STO5. Th eir speculation is that the large carrier concentration is sourced by a thin donor layer of oxygen vacancies inside the STO and just below the LAO. Th ey point out that due to the large dielectric constant of STO the electrons are not so tightly bound to this donor layer and the high mobility could

result from in-plane transport remote from the charged donor layer.

Although the new results suggest oxygen-vacancy-generated carriers in samples grown in oxygen-poor conditions, the opposite case, of fully oxidized structures, off ers another puzzle. Several research groups have found a common lower limit in the carrier density of n-type interfaces at ~1013 cm–2. Th is is more than one order of magnitude less than the 0.5 electron per unit cell derived above. Th e charge abruptly appears when the LAO thickness reaches a critical value6,7. Th is aspect is fully consistent with the picture outlined above in which no charge accumulates in the STO until the polarization catastrophe reaches a critical value. In the experiment by Th iel et al.7, the two-dimensional sheet density and sheet conductance jump to fi xed values that do not change as the thickness of the LAO is increased, and the mobility at 4 K is high at 1,200 cm2 V–1 s–1. If this is due to doping via the polarization discontinuity, such a high mobility indicates the junction and the volume carrying the transport is clean.

Th e magnetic eff ects reported by Brinkman and co-workers are also obtained on fully oxidized samples with a low carrier density (~1013 cm–2). Th e transition to a magnetic state occurs at low temperatures as shown by a hysteresis in the magnetoresistance that appears at 0.3 K. As the density of carriers is small and the evidence indirect, it is not clear what the mechanism responsible for the ordering is and where the moments are. But this result is certainly exciting and will spur more experiments to try to answer these questions.

Th ere are indeed several pieces to this puzzle, and although some of them may be falling into place we cannot say that we have the full picture yet. It is probably safe to say that that there are two diff erent carrier-density regimes, one around 1016 cm–2, and another around 1013 cm–2. In general the thickness of the LAO layer plays a role, but the presence of oxygen vacancies is an important ingredient, as simply annealing bulk STO for very long times at temperatures similar to those used in growth gives rise to oxygen vacancies and electron carriers. Moreover, the mobility values measured at 4 K in the heterojunction samples are in line with those found in thermally reduced STO having volume carrier concentrations in the range of 1018 to 1020 cm–2 (ref. 8). However, if the LAO layer is grown on a SrO-terminated substrate and not on the TiO2 (this gives rise to a p-type interface), the structure is insulating, which strongly suggests that the details of the interface do matter1.

One point to also keep in mind is that there is a very large discontinuity

in the band structure at the junction. More precisely, the valence band is nearly continuous, whereas the conduction band in LAO is about 2.4 eV higher than in STO (Fig. 2). If this discontinuity is somehow modulation-doped, a lot of charge can fi t in there and a sheet carrier concentration in the range of >1013 cm–2 is not out of line. Indeed, the question may be whether higher values of sheet carrier doping can also be obtained while keeping the carriers localized at the interface. More characterization, such as capacitance–voltage measurements and Raman scattering, will help reveal where the carriers are. Th is system can also be extended to more exotic structures, such as wells of fi nite thickness6. Th e extraordinary properties found in oxide system heterostructures may lead to useful devices with novel characteristics. Th e distribution of carriers in such structures will also help clarify what is happening in the experiments discussed here.

References1. Ohtomo, A. & Hwang, H. Y. Nature 427, 423–426 (2004).2. Brinkman, A. et al. Nature Mater. 6, 493–496 (2007).3. Kalabukhov, A. et al. Phys Rev. B 75, 121404 (2007).4. Herranz, G. et al. Phys. Rev. Lett. 98, 216803 (2007).5. Siemons, W. et al. Phys. Rev. Lett. 98, 196802 (2007).6. Huijben, M. et al. Nature Mater. 5, 556–560 (2006).7. Th iel, S. et al. Science 313, 1942–1945 (2006).8. Friederikse, H. P. R. & Hosler, W. R. Phys. Rev. 161, 822–827 (1967)

–eV

x

0

1–

SrO

TiO2

SrO

TiO2

LaO1+

AlO2

LaO1+

LaO1+

AlO2

AlO21–

1–

Figure 2 Layered structure of an n-type interface formed by growing LAO on top of TiO2-terminated STO. In the LAO layers an electric fi eld points upwards, starting from the positively charged LaO layers and terminating in the negatively charged AlO2 layers. The additional electrochemical potential of electrons subject to this fi eld is shown in the graph as a function of the growth direction x. The monotonically increasing potential energy of the electrons at the surface can be avoided if the conduction band in the STO at the interface gets half an electron per unit cell. In fully oxygenated samples, the mobile charge density at the interface is much less than this simple charge-transfer picture suggests, a factor that is not yet understood. A p-type interface is formed when the positively charged LaO layer at the interface is replaced by neutral SrO, and so far no conducting p-channel samples have been made.

ABO3

Figure 1 Ball and stick model of a cubic perovskite ABO3 unit cell. The electronic states near the Fermi energy are formed from oxygen and B-atom molecular orbitals. The A-atom ionizes and transfers two (Sr) or three (La) electrons to the BO3 states. The valence band is principally oxygen in character whereas the conduction band is mainly made up of B-atom orbitals. The bandgap for STO is 3.2 eV, and for LAO is 5.6 eV.

nmat July N&V.indd 474nmat July N&V.indd 474 11/6/07 16:04:1611/6/07 16:04:16